mapping_networks.trainers.MappingTrainer¶

class mapping_networks.trainers.MappingTrainer(model: MappingModel, train_loader: DataLoader[Any], val_loader: DataLoader[Any] | None = None, loss_fn: MappingLoss | BaseLoss | None = None, config: TrainerConfig | None = None, callbacks: list[Callback] | None = None, batch_adapter: BatchAdapter | None = None)¶

Training loop for mapping networks.

The trainer owns optimization only: forward, compute loss, backward, optimizer step, validation, and callbacks. Mapping logic stays in the model/generator/loss subsystems.

Parameters:

model – A MappingModel instance.
train_loader – Training DataLoader.
val_loader – Optional validation DataLoader.
loss_fn – A MappingLoss or BaseLoss instance. If None, defaults to MappingLoss(RegressionLoss()).
config – Trainer configuration. If None, uses defaults.
callbacks – List of Callback instances.
batch_adapter – Adapter for unpacking dataloader batches. Defaults to TupleBatchAdapter.

Example:

trainer = MappingTrainer(
    model=mapping_model,
    train_loader=train_loader,
    loss_fn=MappingLoss(ClassificationLoss()),
    config=TrainerConfig(max_epochs=50, learning_rate=3e-4),
    callbacks=[EarlyStopping(patience=5), MetricLogger()],
)
history = trainer.fit()

__init__(model: MappingModel, train_loader: DataLoader[Any], val_loader: DataLoader[Any] | None = None, loss_fn: MappingLoss | BaseLoss | None = None, config: TrainerConfig | None = None, callbacks: list[Callback] | None = None, batch_adapter: BatchAdapter | None = None) → None¶

Methods

`__init__`(model, train_loader[, val_loader, ...])
`fit`()	Run the full training loop.